Bear in mind that in any network, the network congestion and even the routes will be asymmetrical, so that, for example, streaming a game’s video quality might be satisfactory when the user is passive, but at the point the server becomes less responsive to the player’s clicks or mouse movements, the gaming experience begins to feel less responsive.

“Not only that, but the carrier services may be intentionally asymmetric in terms of bandwidth and directional performance,” explains Thierno Diallo, a product manager and expert in packet technologies for EXFO.

“Your Internet provider may have told you that you have 10 megabits downstream, but only four megabits upstream,” Diallo explained, adding that this is okay because “we consume more data traveling from the network toward us than that which we push up to the network. So when we look at delay loss, and performance in general, it is essential to look at it from a directional perspective. It’s not sufficient to look at round-trip performance.”

Most services are asymmetrical: games, broadcasting, movie streaming, software updates, and business applications, with far more traffic heading downstream toward the customer, and relatively little heading upstream toward the server or cloud provider.

It is critical to take into account this asymmetry. It is not enough to focus on downstream latency. Both upstream and downstream latency must be measured, and measured accurately, in order to understand the overall session end-to-end latency, and to use that data to improve QoE.

GPS-Based Clock Synchronization

Time is relative. When measuring latency in sub-milliseconds, it is essential to ensure that accurate timestamps from synchronized clocks are used when measuring unidirectional packet delays.

The most common approach has been to use accurate time signatures from GPS, which requires extremely accurate timings, down to the microsecond, based on satellite transmissions. Because GPS does not require that network nodes sync with each other, it is the de facto standard; however, the technology is costly to acquire, deploy and manage.

First, not all endpoints possess GPS capability or the capability to be directly connected to a GPS device. Second, distance is a factor: the farther away the GPS, or the more complex the network topology, the less accurate the clock synchronization. Third, even GPS-equipped network devices are not always active or usable because some utilize GPS for networking functionality as opposed to functionality testing.

And last but not least, a major downside of GPS-driven approaches is the fact some carriers measure round-trip packet latency and then divide that measurement by two when calculating an approximation of one-way delay. While this avoids the need for direct clock synchronization, it can be very inaccurate when measuring upstream or downstream performance independently. It can provide a false representation of QoE, and make it difficult to determine where a problem may lie, or how to rectify it.

Active Testing of Latency in Both Directions

“As we get into more transactional services and become a more mobile society,” said EXFO’s Diallo, “the importance of looking at everything from a unidirectional perspective is significantly more important for carriers and enterprise customers. Carriers must gather those metrics in as efficient a manner as possible.” Based on the data produced from those metrics, carriers can gain a real understanding of where delay occurs in the network and subsequently improve traffic engineering to reduce latency.

This can be done without a true end-to-end clock synchronization, even through mobile backhaul, indoor deployments, or mobile devices may lack reliable clocks or access to GPS. Through active testing, it becomes possible to conduct GPS sub-microsecond measurements, even when devices lack access to GPS receivers or other similar clock sync resources.